Community Content Policy

1. Content we do not allow

Do not upload, send, publish, or use Sodium to distribute content that credibly threatens violence; encourages self-harm or harm to others; targets people with hateful or dehumanizing attacks based on protected characteristics; sexually exploits or sexualizes minors; depicts sexual abuse; contains non-consensual intimate imagery; is graphic gore intended to shock or harass; stalks, doxxes, impersonates, or harasses someone; distributes malware, stolen credentials, spam, or unlawful material; or infringes another person’s rights.

Context matters. Ordinary profanity, criticism, quotation, reporting, educational material, medical diagrams, fashion, swimwear, artistic work, and discussion of difficult subjects are not automatically violations. We consider the surrounding context and the likely impact on people, not a single keyword in isolation.

2. What happens when content is submitted

New text and images may be marked pending while checks run. Safe content is approved and becomes available through the feature where it was submitted. Potentially unsafe or uncertain content is quarantined: it is retained privately for review and is not shown to other users. Quarantine is a visibility safeguard, not automatically a strike or a ban.

For text, Sodium uses Cloudflare’s Llama Guard safety classifier and the site’s policy filter. For images, the browser may run NSFWJS for an immediate local check; the server checks a moderation-sized derivative with NudeNet, extracts meaningful text for text moderation, and sends uncertain or reported images to a general vision safety model. GIF-only messages remain outside this image pipeline for now.

The original submitted image is kept separate from the moderation-sized copy. Moderators see only content needed to make a decision, the automated result, relevant history, and reports. Automated systems control visibility; only human moderators issue permanent bans.

3. Human review and possible actions

Moderators may approve content, remove it, issue a warning, apply a temporary restriction, or permanently suspend an account. Decisions are recorded with the moderator, time, action, and reason so the moderation team can review consistency. A confirmed violation can also increase factual account counters such as warnings or recently removed content; these are moderation context, not a public score.

We may take urgent action to prevent imminent harm, preserve evidence, restrict access, or comply with law. We do not promise that every automated decision is correct, and we do not treat a model score as a permanent measure of a person.

4. Reports and appeals

Use the report control next to the relevant message or contact sodiumcollab@gmail.com with the content link, what happened, and any useful context. Do not include private information unless it is necessary to explain the safety issue.

If your content was quarantined or removed in error, use the “held for review” notice when available or email us with the moderation ID and context. A different moderator may review an appeal. Do not repost the same content repeatedly while it is under review, and do not create another account to evade a restriction.

5. Privacy and retention

Moderation records may include submitted content, thumbnails, model outputs, reports, decisions, and limited account history. They are used for safety, appeals, security, and legal obligations, and are access-controlled. See the Privacy Policy for the broader data practices and your available rights. See the Terms and Conditions for the agreement governing use of Sodium.